Modify to check if alpha is in host memory.

[geotseng-amd]

Cherry-pick from #1326, #1335

---

Summary

In this PR, we modified the runtime verifications to check if Alpha is in host memory.

Due to a modification from PR#956, the CPU may try to access device memory through Alpha, which violates the BIOS restrictions.

Assigned Ticket: https://ontrack-internal.amd.com/browse/SWDEV-496511
Caused Ticket: https://ontrack-internal.amd.com/browse/SWDEV-482839
Caused PR: #956

Modifications

In hipBLASLt API

HIPBLASLT_MATMUL_DESC_POINTER_MODE has three modes as the following:

typedef enum {
    HIPBLASLT_POINTER_MODE_HOST = 0,                          /** targets host memory */
    HIPBLASLT_POINTER_MODE_DEVICE = 1,                        /** targets device memory */
    HIPBLASLT_POINTER_MODE_ALPHA_DEVICE_VECTOR_BETA_HOST = 4, /** alpha pointer targets a device memory vector of length equal to the number of rows of matrix D, and beta is a single value in host memory. */
} hipblasLtPointerMode_t;

Only the HIPBLASLT_POINTER_MODE_HOST will ensure alpha is in host memory, so we modified validateMatmulArgs to check pointermode.

In hipBLASLtExt API

Since this ScaleAlphaVec is passed as a parameter through the API, we can simply push a nullptr as pointermode.

[geotseng-amd]

All the local CI tests for PR#1356 are passed.

�[0;32m[----------] �[mGlobal test environment tear-down
�[0;32m[==========] �[m48206 tests from 13 test suites ran. (826691 ms total)
�[0;32m[ PASSED ] �[m48206 tests.
hipBLASLt version: 1000

command line: ./build/release/clients/staging/hipblaslt-test --gtest_output=xml --gtest_color=yes --gtest_repeat=1